Introduction to Shiny, Session 3
Bioinformatics Resource Center - Rockefeller University
http://rockefelleruniversity.github.io/RU_course_template/
brc@rockefeller.edu
Session 3 - Improving app experience - dynamic UIs and user feedback
Recap
- Session 1 -
- Shiny basics
- app layouts
- app theming
- Session 2 -
- Input types
- basic reactivity
- interactive plots
- downloading files from apps
- Session 3 -
- uploading files and conditional UIs
- observers
- publishing apps
- basic debugging
Upload a file
We will pick up wih the app we built in Sessions 1 and 2 (below).
The app contains some nice reactive elements, but this app would be more useful if you could use any file on your computer with differential expression results as opposed to having to change the path in our app to look at a different set of results.
We will modify our app to upload a file.
Upload a file
We can use the fileInput function in the UI to allow the user to input a file. The ‘accept’ argument to limit the type of file the user can try to upload.
ui <- page_fluid(
fileInput("de_file", "Upload a DE file", accept = c(".csv", ".tsv", "xlsx", "xls")), #<<
dataTableOutput(outputId = "all_data"),
)
server <- function(input, output){
de_table_in <- reactive({
rio::import(input$de_file$datapath) %>% dplyr::mutate(negLog10_pval = -log10(pvalue))
})
output$all_data = renderDataTable({
datatable(de_table_in(),
filter = 'top') %>%
formatRound(columns = c("baseMean", "log2FoldChange", "lfcSE", "stat"), digits = 3) %>%
formatSignif(columns = c("pvalue", "padj"), digits = 3)
})
}Upload a file
The help page for fileInput (?fileInput) explains that once a file is loaded, then the value returned to the server is a data frame, and one of the columns is the path to the temporary file path where Shiny is holding the file.
This path is used below in the de_table_in reactive expression to read in the dataframe.
ui <- page_fluid(
fileInput("de_file", "Upload a DE file", accept = c(".csv", ".tsv", "xlsx", "xls")),
dataTableOutput(outputId = "all_data"),
)
server <- function(input, output){
de_table_in <- reactive({
rio::import(input$de_file$datapath) %>% #<<
dplyr::mutate(negLog10_pval = -log10(pvalue))
})
output$all_data = renderDataTable({
datatable(de_table_in(),
filter = 'top') %>%
formatRound(columns = c("baseMean", "log2FoldChange", "lfcSE", "stat"), digits = 3) %>%
formatSignif(columns = c("pvalue", "padj"), digits = 3)
})
}
The req() function
In the previous UI, the user sees an error until a file is uploaded. This is because the file path is NULL and the rio::import function throws an error.
Shiny has a handy function req that can be added to a reactive context and the reactive or output function won’t run if the value passed to req is NULL. We modify the reactive in the server function that reads in the table.
server <- function(input, output){
de_table_in <- reactive({
req(input$de_file) #<<
rio::import(input$de_file$datapath) %>%
dplyr::mutate(negLog10_pval = -log10(pvalue))
})
output$all_data = renderDataTable({
datatable(de_table_in(),
filter = 'top') %>%
formatRound(columns = c("baseMean", "log2FoldChange", "lfcSE", "stat"), digits = 3) %>%
formatSignif(columns = c("pvalue", "padj"), digits = 3)
})
}
Add upload to main app - UI
ui <- page_navbar(
title = "RNAseq tools",
theme = custom_theme,
nav_panel(
title = "DE Analysis",
layout_sidebar(
sidebar = sidebar(
width = 300,
# >>>>>>>>>
fileInput("de_file", "Upload a DE file", accept = c(".csv", ".tsv", "xlsx", "xls")),
# >>>>>>>>>
numericInput("padj_filter", label = "Cutoff for padj:", value = 0.05, min = 0, max = 1, step = 0.005),
numericInput("lfc_filter", label = "Cutoff for log2 FC:", value = 1, min = 0, step = 0.1),
actionButton("de_filter", "Apply filter"),
br(),
br(),
value_box(title = "Number of genes that go up:", value = textOutput("num_up"),
showcase = icon("arrow-up"),
theme = value_box_theme(bg = "#22b430")),
value_box(title = "Number of genes that go down:", value = textOutput("num_down"),
showcase = icon("arrow-down"),
theme = value_box_theme(bg ="#c34020" ))
),
layout_columns(
navset_card_tab(
title = "DE result tables",
nav_panel(card_header("DEGs"), dataTableOutput(outputId = "de_data")),
nav_panel(card_header("All genes"), dataTableOutput(outputId = "all_data"))
),
card(card_header("MA plot"),
plotlyOutput("ma_plot"),
downloadButton("download_ma_plot", "Download MA plot", style = "width:40%;")),
card(card_header("Volcano plot"),
plotlyOutput("volcano_plot"),
downloadButton("download_volcano_plot", "Download volcano plot", style = "width:40%;")),
col_widths = c(12,6,6), row_heights = c("750px", "500px")
)
)
),
nav_panel(title = "Next steps","The next step in our analysis will be..."),
nav_spacer(),
nav_menu(title = "Links",
align = "right",
nav_item(tags$a(shiny::icon("chart-simple"), "RU BRC - Learn more!", href = "https://rockefelleruniversity.github.io/",target = "_blank"))
)
)Add upload to main app - server
The filtered table reactive and plot reactives use this table to apply the filtering cut offs, so we change these reactives to use this table and add de_table_in() to bindEvent so that they are updated when a new dataset is uploaded.
server = function(input, output) {
# >>>>>>>>>>>>>>>>>>>>>>>>
de_table_in <- reactive({
req(input$de_file)
rio::import(input$de_file$datapath) %>% dplyr::mutate(negLog10_pval = -log10(pvalue))
})
# >>>>>>>>>>>>>>>>>>>>>>>>
output$all_data = renderDataTable({
datatable(de_table_in(), # >>>>>>>>>>>>>>>>>>>>>>>>
filter = 'top') %>%
formatRound(columns = c("baseMean", "log2FoldChange", "lfcSE", "stat"), digits = 3) %>%
formatSignif(columns = c("pvalue", "padj"), digits = 3)
})
filtered_de <- reactive({
de_table_in() %>% # >>>>>>>>>>>>>>>>>>>>>>>>
dplyr::filter(padj < input$padj_filter & abs(log2FoldChange) > input$lfc_filter)
}) %>%
bindEvent(input$de_filter, de_table_in(), ignoreNULL = FALSE) # >>>>>>>>>>>>>>>>>>>>>>>>
output$de_data = renderDataTable({
datatable(filtered_de(),
filter = 'top') %>%
formatRound(columns = c("baseMean", "log2FoldChange", "lfcSE", "stat"), digits = 3) %>%
formatSignif(columns = c("pvalue", "padj"), digits = 3)
})
num_up_genes <- reactive(filtered_de() %>% dplyr::filter(log2FoldChange > 0 & padj < 0.05) %>% nrow)
num_down_genes <- reactive(filtered_de() %>% dplyr::filter(log2FoldChange < 0 & padj < 0.05) %>% nrow)
output$num_up <- renderText(num_up_genes())
output$num_down <- renderText(num_down_genes())
ma_plot_reac <- reactive({
de_table_in() %>% # >>>>>>>>>>>>>>>>>>>>>>>>
dplyr::mutate(sig = ifelse(padj < input$padj_filter & abs(log2FoldChange) > input$lfc_filter, "DE", "Not_DE")) %>%
ggplot(aes(x = baseMean, y = log2FoldChange, color = sig, label = Symbol)) + geom_point() +
scale_x_log10() + scale_color_manual(name = "DE status", values = c("red", "grey")) +
xlab("baseMean (log scale)") + theme_bw()
}) %>%
bindEvent(input$de_filter, de_table_in(), ignoreNULL = FALSE) # >>>>>>>>>>>>>>>>>>>>>>>>
output$ma_plot = renderPlotly({
ggplotly(ma_plot_reac())
})
volcano_plot_reac <- reactive({
de_table_in() %>% # >>>>>>>>>>>>>>>>>>>>>>>>
dplyr::mutate(sig = ifelse(padj < input$padj_filter & abs(log2FoldChange) > input$lfc_filter, "DE", "Not_DE")) %>%
ggplot(aes(x = log2FoldChange, y = negLog10_pval, color = sig)) +
geom_point() +
scale_color_manual(name = "DE status", values = c("red","grey")) + theme_bw()
}) %>%
bindEvent(input$de_filter, de_table_in(), ignoreNULL = FALSE) # >>>>>>>>>>>>>>>>>>>>>>>>
output$volcano_plot = renderPlotly({
ggplotly(volcano_plot_reac())
})
output$download_ma_plot <- downloadHandler(
filename = function() {
"maplot.pdf"
},
content = function(file) {
ggsave(filename = file, plot = ma_plot_reac())
}
)
output$download_volcano_plot <- downloadHandler(
filename = function() {
"volcanoplot.pdf"
},
content = function(file) {
ggsave(filename = file, plot = volcano_plot_reac())
}
)
}
Using uiOutput/renderUI functions
]
We still have empty boxes when the app starts even though we don’t have any data to fill that space. There are also buttons that don’t do anything because there is no data yet. This is likely to be confusing for a user.
A nice way to deal with this is the uiOutput function, which allows you to change the user interface after the app is running based on other inputs or code in the server.
So far our user interface is set up at the start and while the contents might change based on other reactives, we haven’t been able to make new inputs or outputs after the app has been started. ]
Using uiOutput/renderUI functions
]
We could improve the flow of our app by making the filter inputs in the sidebar only appear once a user has loaded in a differential table.
These filter inputs aren’t relevant until the data is loaded, so we will only make them appear once the de_table_in() value is a dataframe, suggesting a file has been loaded and a table successfully read in. ]
Using uiOutput/renderUI functions
The inputs for applying filters to our differential table are replaced with a uiOutput function call with an ID used in the output object in the server function This holds a location within the UI for us to eventually fill with server code.
ui <- page_fluid(
fileInput("de_file", "Upload a DE file", accept = c(".csv", ".tsv", "xlsx", "xls")),
uiOutput("sidebar_filters_UI"), #<<
)
server <- function(input, output){
de_table_in <- reactive({
req(input$de_file)
rio::import(input$de_file$datapath) %>% dplyr::mutate(negLog10_pval = -log10(pvalue))
})
output$sidebar_filters_UI <- renderUI({
req(de_table_in())
div(numericInput("padj_filter", label = "Cutoff for padj:", value = 0.05, min = 0, max = 1, step = 0.001),
numericInput("lfc_filter", label = "Cutoff for log2 FC:", value = 1, min = 0, step = 0.1),
actionButton("de_filter", "Apply filter"))
})
}Using uiOutput/renderUI functions
These inputs are moved to the server within an output object paired with renderUI and are conditional on de_table_in() being a dataframe.
library(rio)
ui <- page_fluid(
fileInput("de_file", "Upload a DE file", accept = c(".csv", ".tsv", "xlsx", "xls")),
uiOutput("sidebar_filters_UI"),
)
server <- function(input, output){
de_table_in <- reactive({
req(input$de_file)
rio::import(input$de_file$datapath) %>% dplyr::mutate(negLog10_pval = -log10(pvalue))
})
output$sidebar_filters_UI <- renderUI({ #<<
req(de_table_in()) #<<
div(numericInput("padj_filter", label = "Cutoff for padj:", value = 0.05, min = 0, max = 1, step = 0.001),#<<
numericInput("lfc_filter", label = "Cutoff for log2 FC:", value = 1, min = 0, step = 0.1),#<<
actionButton("de_filter", "Apply filter"))#<<
})#<<
}The div() function
You might have noticed in the server function that we wrapped the three UI elements within renderUI in a div() function call.
We do this because renderUI will only return a single UI element. The R function div will group mutliple HTML elements into one object that is compatible with renderUI.
If you’ve ever looked at HTML code, you’ll notice it’s containerized
into chunks divided by <div> tags, and div
is one of the HTML helper functions in R, grouping mutliple elements
into one of these containers.
server <- function(input, output){
de_table_in <- reactive({
req(input$de_file)
rio::import(input$de_file$datapath) %>% dplyr::mutate(negLog10_pval = -log10(pvalue))})
output$sidebar_filters_UI <- renderUI({
req(de_table_in())
div( #<<
numericInput("padj_filter", label = "Cutoff for padj:", value = 0.05, min = 0, max = 1, step = 0.001),
numericInput("lfc_filter", label = "Cutoff for log2 FC:", value = 1, min = 0, step = 0.1),
actionButton("de_filter", "Apply filter")
) #<<
})
}
Using uiOutput/renderUI functions
]
We will also hide the tables and plots since they are of no use until a file is uploaded. Empty elements can confuse the user and make it seem like something is wrong. ]
Using uiOutput/renderUI functions
Conditional UIs can also take advantage of more complex if statements to determine what is shown. In the example below, if no data frame is loaded, then we output a message for the user and once data is loaded, the table is shown.
ui <- page_fluid(
fileInput("de_file", "Upload a DE file", accept = c(".csv", ".tsv", "xlsx", "xls")),
uiOutput("all_data_UI") #<<
)
server <- function(input, output){
de_table_in <- reactive({
req(input$de_file)
rio::import(input$de_file$datapath) %>% dplyr::mutate(negLog10_pval = -log10(pvalue))})
output$all_data_UI <- renderUI({ #<<
if(is.null(input$de_file)) { #<<
div("You must load data!", style = "color: #273449; font-weight: bold;") #<<
}else if(!is.null(de_table_in())){ #<<
navset_card_tab(nav_panel(card_header("All genes"), dataTableOutput(outputId = "all_data"))) #<<
} #<<
}) #<<
output$all_data = renderDataTable(datatable(de_table_in()))
}
conditionalPanel
The conditionalPanel function can also be used for dynamic display of UI elements. It allows the defining of mutliple UI options in the UI object based on another input, so uiOutput/renderUI is not necessary.
conditionalPanel works well when you have multiple options that can be displayed downstream of a TRUE/FALSE input like a check box or radio button, or something with discrete and known outputs, like dropdown lists.
conditionalPanel
The two important arguments are ‘condition’, and if this expression evaluates as TRUE, then the UI elements within the function will be displayed.
Note that the conditional statement takes a different form then we are used to, its actually a JavaScript expression. These can get complex, but generally you will be using as shown below: ‘input.inputID is equal (==) ir not equal (!=) to a value (eg ’1’ for TRUE or ‘0’ for FALSE)
We won’t use this function in our app, but here is a simple example of conditionalPanel using a nested structure which creates a cascading set of panels.
ui <- page_fluid(
checkboxInput("question", "Do you want to use my app?"),
conditionalPanel(condition = "input.question == '1'",
selectInput("experiment", "What kind of experiment is this?",
choices = c("", "RNAseq", "ATACseq")),
conditionalPanel(condition = "input.experiment == 'RNAseq'",
fileInput("file_in", "Great!, upload your RNAseq file:")),
conditionalPanel(condition = "input.experiment == 'ATACseq'",
"Sorry, but this app won't help you"))
)
server = function(input, output){}Launch conditionalPanel app
]
- When the user checks the box, the dropdown appears
- condtion = “input.question == ‘1’”.
- If the user selects RNAseq, then we let them upload a file.
- condition = “input.experiment == ‘RNAseq’”
- If the user selects ATACseq, we give a message saying this app isn’t
for them.
- condition = “input.experiment == ‘ATACseq’” ]
conditionalPanel vs uiOutput
- conditionalPanel works nicely when the UI that you want
hidden will not change how it looks based on things that are happening
in the server function.
- In our example, we know exactly how the selectInput, fileInput, and text will look. They are fixed, but we just want them hidden and dependent on inputs that are already in the UI object.
- Once you start depending on server side logic to render your UI
elements, then uiOutput is likely the better option.
- In our example, we need the reactive that reads the table to return a dataframe before we can display the table. This needs work in the server function to figure this out, so we use uiOutput.
- In more complicated apps that are beyond the scope of this course, you can dynamically make varying numbers of inputs based on user input, and uiOutput is good for this as well.
Update main app
Recap of changes: - use uiOutput/renderUI to make the filter inputs and button from the sidebar conditional on the table being uploaded - use uiOutput/renderUI to display a message if there is no datapath loaded and only show the DE table one a valid table is read into the app.
Update main app - UI
ui <- page_navbar(
title = "RNAseq tools",
theme = custom_theme,
nav_panel(
title = "DE Analysis",
layout_sidebar(
sidebar = sidebar(
width = 300,
fileInput("de_file", "Upload a DE file", accept = c(".csv", ".tsv", "xlsx", "xls")),
uiOutput("sidebar_filters_UI") # >>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>
),
uiOutput("table_plots_UI"), # >>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>
)
),
nav_panel(title = "Next steps","The next step in our analysis will be..."),
nav_spacer(),
nav_menu(title = "Links",
align = "right",
nav_item(tags$a(shiny::icon("chart-simple"), "RU BRC - Learn more!", href = "https://rockefelleruniversity.github.io/",target = "_blank"))
)
)Update main app - server
server = function(input, output) {
de_table_in <- reactive({
req(input$de_file)
file_in <- rio::import(input$de_file$datapath) %>% dplyr::mutate(negLog10_pval = -log10(pvalue))
})
# >>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>
output$table_plots_UI <- renderUI({
if(is.null(input$de_file)) {
layout_columns("No data has been loaded! Upload a DE table with the following columns: 'baseMean', 'log2FoldChange', 'lfcSE', 'stat', 'pvalue', 'padj'", style = "color: #273449; font-weight: bold;")
}else if(!is.null(de_table_in())){
layout_columns(
navset_card_tab(
title = "DE result tables",
nav_panel(card_header("DEGs"), dataTableOutput(outputId = "de_data")),
nav_panel(card_header("All genes"), dataTableOutput(outputId = "all_data"))
),
card(card_header("MA plot"),
plotlyOutput("ma_plot"),
downloadButton("download_ma_plot", "Download MA plot", style = "width:40%;")),
card(card_header("Volcano plot"),
plotlyOutput("volcano_plot"),
downloadButton("download_volcano_plot", "Download volcano plot", style = "width:40%;")),
col_widths = c(12,6,6), row_heights = c("750px", "500px")
)
}
})
# >>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>
# >>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>
output$sidebar_filters_UI <- renderUI({
req(de_table_in())
div(
"DE filters",
numericInput("padj_filter", label = "Cutoff for padj:", value = 0.05, min = 0, max = 1, step = 0.001),
numericInput("lfc_filter", label = "Cutoff for log2 FC:", value = 1, min = 0, step = 0.1),
actionButton("de_filter", "Apply filter"),
br(),
br(),
value_box(title = "Number of genes that go up:", value = textOutput("num_up"),
showcase = icon("arrow-up"),
theme = value_box_theme(bg = "#22b430")),
value_box(title = "Number of genes that go down:", value = textOutput("num_down"),
showcase = icon("arrow-down"),
theme = value_box_theme(bg ="#c34020" ))
)
})
# >>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>
output$all_data = renderDataTable({
datatable(de_table_in(), filter = 'top') %>%
formatRound(columns = c("baseMean", "log2FoldChange", "lfcSE", "stat"), digits = 3) %>%
formatSignif(columns = c("pvalue", "padj"), digits = 3)
})
filtered_de <- reactive({
de_table_in() %>%
dplyr::filter(padj < input$padj_filter & abs(log2FoldChange) > input$lfc_filter)
}) %>%
bindEvent(input$de_filter, de_table_in(), ignoreNULL = FALSE)
num_up_genes <- reactive(filtered_de() %>% dplyr::filter(log2FoldChange > 0 & padj < 0.05) %>% nrow)
num_down_genes <- reactive(filtered_de() %>% dplyr::filter(log2FoldChange < 0 & padj < 0.05) %>% nrow)
output$num_up <- renderText(num_up_genes())
output$num_down <- renderText(num_down_genes())
output$de_data = renderDataTable({
datatable(filtered_de(),
filter = 'top') %>%
formatRound(columns = c("baseMean", "log2FoldChange", "lfcSE", "stat"), digits = 3) %>%
formatSignif(columns = c("pvalue", "padj"), digits = 3)
})
ma_plot_reac <- reactive({
de_table_in() %>%
dplyr::mutate(sig = ifelse(padj < input$padj_filter & abs(log2FoldChange) > input$lfc_filter, "DE", "Not_DE")) %>%
ggplot(aes(x = baseMean, y = log2FoldChange, color = sig, label = Symbol)) + geom_point() +
scale_x_log10() + scale_color_manual(name = "DE status", values = c("red", "grey")) +
xlab("baseMean (log scale)") + theme_bw()
}) %>%
bindEvent(input$de_filter, de_table_in(), ignoreNULL = FALSE)
output$ma_plot = renderPlotly({
ggplotly(ma_plot_reac())
})
volcano_plot_reac <- reactive({
de_table_in() %>%
dplyr::mutate(sig = ifelse(padj < input$padj_filter & abs(log2FoldChange) > input$lfc_filter, "DE", "Not_DE")) %>%
ggplot(aes(x = log2FoldChange, y = negLog10_pval, color = sig, label = Symbol)) +
geom_point() +
scale_color_manual(name = "DE status", values = c("red","grey"),) + theme_bw()
}) %>%
bindEvent(input$de_filter, de_table_in(), ignoreNULL = FALSE)
output$volcano_plot = renderPlotly({
ggplotly(volcano_plot_reac())
})
# download handlers for plots
output$download_ma_plot <- downloadHandler(
filename = function() {
"maplot.pdf"
},
content = function(file) {
ggsave(filename = file, plot = ma_plot_reac())
}
)
output$download_volcano_plot <- downloadHandler(
filename = function() {
"volcanoplot.pdf"
},
content = function(file) {
ggsave(filename = file, plot = volcano_plot_reac())
}
)
}
User Feedback and observers
Handling invalid input files
A common problem when allowing an input file is the likelihood a user uploads a file that causes an error in the app. Here we are looking for a table with speific columns, so we should confirm that the file is valid.
Use validate function to check input file
There are a few ways to do this in Shiny, and we will introduce a new Shiny function to handle this, the validate function.
Validate prevents the alarming red error messages that are unhelpful to the user. This function can be used within a reactive expression, and the validation test is often called within a need function call.
Need takes an expression to evaluate, and if it is FALSE, then it will display a string provided in the ‘message’ argument in any output that depends on this reactive.
Use validate function to check input file
We use validate in the server function to check for the key columns in the table as we know that not having these columns will cause a downstream error in the app.
ui <- page_fluid(
fileInput("de_file", "Upload a DE file", accept = c(".csv", ".tsv", "xlsx", "xls")),
uiOutput("all_data_UI")
)
server <- function(input, output){
de_table_in <- reactive({
req(input$de_file)
file_in <- rio::import(input$de_file$datapath)
validate(need(expr = all(c("baseMean", "log2FoldChange", "lfcSE", "stat", "pvalue", "padj") %in% colnames(file_in)), #<<
message = "You must have the following columns: 'baseMean', 'log2FoldChange', 'lfcSE', 'stat', 'pvalue', 'padj'")) #<<
file_in %>% dplyr::mutate(negLog10_pval = -log10(pvalue))
})
output$all_data_UI <- renderUI({
if(is.null(input$de_file)) {
div("You must load data!", style = "color: #273449; font-weight: bold;")
}else if(!is.null(de_table_in())){
navset_card_tab(nav_panel(card_header("All genes"), dataTableOutput(outputId = "all_data")))
}
})
output$all_data = renderDataTable(datatable(de_table_in()))
}
Observers
Sometimes we might want our app to react to a change in an input, but we don’t need to return a value like reactive or we don’t need to change one of the outputs. Maybe we want to write to a database when a button is clicked, or notify the user that something has happened.
Often the observe function is used for this purpose, to perform a side effect when an input changes.
Like the reactive function or an output, observe creates a reactive context that takes dependencies on inputs.
Though unlike a reactive expression, an observer does not return a value and is eager in its evaluation, meaning that it will evaluate the code every time an input it depends on changes.
Observers - notification for user
We will add a nice message for the user to notify them that a new data set has been loaded.
To do this we use the Shiny function showNotification. This takes text that will be the message, a duration in seconds for the notification to remain open, and a ‘type’ argument, which will control the color. We set ‘duration’ to be NULL, which means the user will have to click to close the notification, guaranteeing they will see it.
This function is within an observe function call in the server and takes a dependency on the input table with bindEvent. Notice we don’t set the result to be a variable because an observer returns nothing, it just runs the code it contains.
Observers - notification for user
ui <- page_fluid(
fileInput("de_file", "Upload a DE file", accept = c(".csv", ".tsv", "xlsx", "xls")),
uiOutput("all_data_UI"))
server <- function(input, output){
de_table_in <- reactive({
req(input$de_file)
file_in <- rio::import(input$de_file$datapath)
})
observe({ #<<
showNotification("A new table has been loaded into the app!", duration = NULL, type = "message") #<<
}) %>% #<<
bindEvent(de_table_in()) #<<
output$all_data_UI <- renderUI({
if(is.null(input$de_file)) {
div("Load data!", style = "color: #273449; font-weight: bold;")
}else{ navset_card_tab(nav_panel(card_header("All genes"), dataTableOutput(outputId = "all_data"))) }
})
output$all_data = renderDataTable(datatable(de_table_in()))
}
Modals for messages to user
Another commonly used element paired with observe is an modal, which is essentially a pop up window that can simply display a message to the user, or contain additional UI elements.
In the simple app below, we use the showModal and modalDialog functions within an observer that takes the button from the UI as a dependency.
ui <- page_fluid(
actionButton("show_modal",
"Click me for modal!")
)
server <- function(input, output){
observe({
showModal(
modalDialog(
title = "The button was clicked!",
easyClose = TRUE,
footer = modalButton("Dismiss")))
}) %>% bindEvent(input$show_modal)
}]
]
Modals to confirm user action
Modals are often used to show additional UI elements, including user inputs, plots or tables. This can be especially useful to confirm actions that make permanent changes, such as writing to a database.
Modals can also be helpful to gather more information from the user for a particular action, and in the following app we add UI elements to the modal to get a file name from the user before downloading the file.
UI:
Modals to confirm user action
In the server we add the UI elements to the modalDialog function and the use the name from the modal in the file name for download.
server <- function(input, output){
volcano_plot_reac <- reactive(ggplot(de_table, aes(x = log2FoldChange, y = negLog10_pval, text = Symbol)) + geom_point() + theme_bw())
output$volcano_plot = renderPlot(ggplot(de_table, aes(x = log2FoldChange, y = negLog10_pval, text = Symbol)) + geom_point() + theme_bw())
observe({
showModal(
modalDialog(title = "Are you sure you want to download this plot? If so, provide a name for the file.",
textInput("file_name", "Name for file (no extension):", value = "volcano_modal"), #<<
downloadButton("confirm_download", "Download plot"), #<<
modalButton("Don't download plot"),
easyClose = TRUE, footer = NULL))
}) %>% bindEvent(input$volcano_download_modal)
output$confirm_download <- downloadHandler(
filename = function() {
paste0(input$file_name, ".pdf") #<<
}, content = function(file) {
ggsave(filename = file, plot = volcano_plot_reac())
})
}
Other user feedback packages
- shinyFeedback or shinyValidate to validate individual inputs
- shinybusy for busy spinners
- shinycssloaders for alerts and busy spinners
- Progress bars in shiny
- Progress bars with waiter pacakge
Update main app
Recap of changes: - add validate + need to the reactive expression where we read in the table from the user to make sure a valid input file was used - add notification to user that a table has been uploaded - add modal dialog to confirm download and get filename
Update main app - UI
The UI has not changed since we previously modified it, but it’s shown below.
ui <- page_navbar(
title = "RNAseq tools",
theme = custom_theme,
nav_panel(
title = "DE Analysis",
layout_sidebar(
sidebar = sidebar(
width = 300,
fileInput("de_file", "Upload a DE file", accept = c(".csv", ".tsv", "xlsx", "xls")),
uiOutput("sidebar_filters_UI")
),
uiOutput("table_plots_UI"),
)
),
nav_panel(title = "Next steps","The next step in our analysis will be..."),
nav_spacer(),
nav_menu(title = "Links",
align = "right",
nav_item(tags$a(shiny::icon("chart-simple"), "RU BRC - Learn more!", href = "https://rockefelleruniversity.github.io/",target = "_blank"))
)
)Update app with user feedback - server
server = function(input, output) {
de_table_in <- reactive({
req(input$de_file)
file_in <- rio::import(input$de_file$datapath)
# >>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>
validate(
need(expr = all(c("baseMean", "log2FoldChange", "lfcSE", "stat", "pvalue", "padj") %in% colnames(file_in)),
message = "You must have the following columns: 'baseMean', 'log2FoldChange', 'lfcSE', 'stat', 'pvalue', 'padj'")
)
# >>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>
file_in %>% dplyr::mutate(negLog10_pval = -log10(pvalue))
})
# >>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>
observe({
showNotification("A new table has been loaded into the app!", duration = NULL, type = "message")
}) %>%
bindEvent(de_table_in())
# >>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>
output$sidebar_filters_UI <- renderUI({
req(de_table_in())
div(
"DE filters",
numericInput("padj_filter", label = "Cutoff for padj:", value = 0.05, min = 0, max = 1, step = 0.001),
numericInput("lfc_filter", label = "Cutoff for log2 FC:", value = 1, min = 0, step = 0.1),
actionButton("de_filter", "Apply filter"),
br(),
br(),
value_box(title = "Number of genes that go up:", value = textOutput("num_up"),
showcase = icon("arrow-up"),
theme = value_box_theme(bg = "#22b430")),
value_box(title = "Number of genes that go down:", value = textOutput("num_down"),
showcase = icon("arrow-down"),
theme = value_box_theme(bg ="#c34020" ))
)
})
output$table_plots_UI <- renderUI({
if(is.null(input$de_file)) {
layout_columns("No data has been loaded! Upload a DE table with the following columns: 'baseMean', 'log2FoldChange', 'lfcSE', 'stat', 'pvalue', 'padj'", style = "color: #273449; font-weight: bold;")
}else if(!is.null(de_table_in())){
layout_columns(
navset_card_tab(
title = "DE result tables",
nav_panel(card_header("DEGs"), dataTableOutput(outputId = "de_data")),
nav_panel(card_header("All genes"), dataTableOutput(outputId = "all_data"))
),
card(card_header("MA plot"),
plotlyOutput("ma_plot"),
actionButton("ma_download_modal", "Download MA plot", width = "40%")),# >>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>
card(card_header("Volcano plot"),
plotlyOutput("volcano_plot"),
actionButton("volcano_download_modal", "Download volcano plot", width = "40%")), # >>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>
col_widths = c(12,6,6), row_heights = c("750px", "500px")
)
}
})
# >>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>
observe({
showModal(
modalDialog(title = "Are you sure you want to download this plot? If so, provide a name for the file.",
textInput("file_name_ma", "Name for file (no extension):", value = "ma_plot"),
downloadButton("download_ma_plot", "Download MA plot", style = "width:40%;"),
br(),
br(),
modalButton("Don't download plot"),
easyClose = TRUE, footer = NULL))
}) %>%
bindEvent(input$ma_download_modal)
# >>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>
# >>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>
observe({
showModal(
modalDialog(title = "Are you sure you want to download this plot? If so, provide a name for the file.",
textInput("file_name_volcano", "Name for file (no extension):", value = "volcano_plot"),
downloadButton("download_volcano_plot", "Download plot", style = "width:40%;"),
br(),
br(),
modalButton("Don't download plot"),
easyClose = TRUE, footer = NULL))
}) %>%
bindEvent(input$volcano_download_modal)
# >>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>
output$all_data = renderDataTable({
datatable(de_table_in(), filter = 'top') %>%
formatRound(columns = c("baseMean", "log2FoldChange", "lfcSE", "stat"), digits = 3) %>%
formatSignif(columns = c("pvalue", "padj"), digits = 3)
})
filtered_de <- reactive({
de_table_in() %>%
dplyr::filter(padj < input$padj_filter & abs(log2FoldChange) > input$lfc_filter)
}) %>%
bindEvent(input$de_filter, de_table_in(), ignoreNULL = FALSE)
num_up_genes <- reactive(filtered_de() %>% dplyr::filter(log2FoldChange > 0 & padj < 0.05) %>% nrow)
num_down_genes <- reactive(filtered_de() %>% dplyr::filter(log2FoldChange < 0 & padj < 0.05) %>% nrow)
output$num_up <- renderText(num_up_genes())
output$num_down <- renderText(num_down_genes())
output$de_data = renderDataTable({
datatable(filtered_de(),
filter = 'top') %>%
formatRound(columns = c("baseMean", "log2FoldChange", "lfcSE", "stat"), digits = 3) %>%
formatSignif(columns = c("pvalue", "padj"), digits = 3)
})
ma_plot_reac <- reactive({
de_table_in() %>%
dplyr::mutate(sig = ifelse(padj < input$padj_filter & abs(log2FoldChange) > input$lfc_filter, "DE", "Not_DE")) %>%
ggplot(aes(x = baseMean, y = log2FoldChange, color = sig, label = Symbol)) + geom_point() +
scale_x_log10() + scale_color_manual(name = "DE status", values = c("red", "grey")) +
xlab("baseMean (log scale)") + theme_bw()
}) %>%
bindEvent(input$de_filter, de_table_in(), ignoreNULL = FALSE)
output$ma_plot = renderPlotly({
ggplotly(ma_plot_reac())
})
volcano_plot_reac <- reactive({
de_table_in() %>%
dplyr::mutate(sig = ifelse(padj < input$padj_filter & abs(log2FoldChange) > input$lfc_filter, "DE", "Not_DE")) %>%
ggplot(aes(x = log2FoldChange, y = negLog10_pval, color = sig, label = Symbol)) +geom_point() +
scale_color_manual(name = "DE status", values = c("red","grey"),) +theme_bw()
}) %>%
bindEvent(input$de_filter, de_table_in(), ignoreNULL = FALSE)
output$volcano_plot = renderPlotly({
ggplotly(volcano_plot_reac())
})
output$download_ma_plot <- downloadHandler(
filename = function() {
paste0(input$file_name_ma, ".pdf")
},
content = function(file) {
ggsave(filename = file, plot = ma_plot_reac())
}
)
output$download_volcano_plot <- downloadHandler(
filename = function() {
paste0(input$file_name_volcano, ".pdf")
},
content = function(file) {
ggsave(filename = file, plot = volcano_plot_reac())
}
)
}
reactiveValues and reactiveVal
We will briefly introduce two more types of reactive variables that can be useful for certain use cases: reactiveVal and reactiveValues.
These differ from other reactive elements becuase they are mutable and can be updated in different parts of your app.
- reactiveVal - a single reactive variable +
counter <- reactiveVal(0)- create a reactiveVal counter with a value of zero +counter <- counter(counter+1)- increase the value of counter with the current value of counter +1 (so now = 1) - reactiveValues - a list-like object that can store
many reactive values
rv <- reactiveValues(counter = 0, tracker = FALSE)- create a reactiveValues object with two elements, rv\(counter* and *rv\)tracker.rv$counter <- rv$counter + 1- increase value of only rv\(counter* to 1, while *rv\)tracker is unchanged.rv$tracker <- TRUE- change value of rv$tracker to TRUE
reactiveValues and reactiveVal
There are many use cases for these two types of variables, and often it’s more useful in more complicated apps, so we won’t get into too much detail here.
Whether you use reactiveVal or reactiveValues is often a matter of personal preference as they can oftne be used to accomplish the same thing.
Here are a couple of examples of how they may be used:
- Tracking the state of the app, for example to know how many times something has happened (a button click) or know if something key has happened.
- If a reactive value needs to be updated in different ways by two
distinct dependencies, reactiveValues and reactiveVal can be used.
- See the next slide for an example of this.
reactiveValues example
We have a simple app where the user is generating a list of genes of interest as they go through the app. We have a button to add a gene to this list and another to remove a gene from this list.
A reactiveVal or reactiveValues object makes it easy to modify this gene list, and importantly, they are reactive and can be used in reactive contexts (see server function)
reactiveValues example
We make two separate observers that make distinct modifications to the reactiveValues object rv$genes.
server <- function(input, output){
# initialize the reactiveValues object
rv <- reactiveValues(genes = character()) #<<
output$all_data = renderDataTable(datatable(de_table, selection = "single", filter = 'top'))
observe({
selected_gene <- de_table[input$all_data_rows_selected, "Symbol"]
rv$genes <- c(rv$genes, selected_gene) #<<
}) %>% bindEvent(input$add_gene) # observer when 'add_gene' button is pressed
observe({
selected_gene <- de_table[input$all_data_rows_selected, "Symbol"]
rv$genes <- rv$genes[!rv$genes == selected_gene] #<<
}) %>% bindEvent(input$remove_gene) # observer when 'remove_gene' button is pressed
output$selected_gene_info <- renderText({
paste0("The selected genes are ", paste(rv$genes, collapse = ", "))
})
}reactiveValues example
Then because rv\(genes* is reactive, using it within *output\)selected_gene_info will trigger that output to change as we add and remove genes.
server <- function(input, output){
rv <- reactiveValues(genes = character())
output$all_data = renderDataTable(datatable(de_table, selection = "single", filter = 'top'))
observe({
selected_gene <- de_table[input$all_data_rows_selected, "Symbol"]
rv$genes <- c(rv$genes, selected_gene)
}) %>% bindEvent(input$add_gene)
observe({
selected_gene <- de_table[input$all_data_rows_selected, "Symbol"]
rv$genes <- rv$genes[!rv$genes == selected_gene]
}) %>% bindEvent(input$remove_gene)
output$selected_gene_info <- renderText({
paste0("The selected genes are ", paste(rv$genes, collapse = ", ")) #<<
})
}
Publishing apps on shinyapps.io
While it may be useful to simply have a Shiny app on your computer that you can run and analyze data locally, you might also want to publish the app in order to share with others, or allow you to access it anywhere.
Posit (aka RStudio) provides the opportunity to deploy apps for free on shinyapps.io, which is nicely integrated into RStudio. The free version allows for a limited number of apps and not much memory, but is a good place to get started. We will go through a simple deployment.
We first need to install and load the rsconnect package.
Publishing apps on shinyapps.io
It is then necessary to make an account on shinyapps.io, and then use the rsconnect package to connect RStudio to the shinyapps.io account.
First the token from shinyapps.io needs to be retrieved:
Publishing apps on shinyapps.io
After running the command copied from shinyapps.io that includes the token and secret, we can then publish our app.
If you have a valid app file open, right next to the ‘Run App’ button, there is another button that allows you to publish the app. The shinyapps.io account that you just linked should be there for deployment.
Publishing apps on shinyapps.io
The rsconnect package will then bundle the app and any packages the app uses. After some time, the log in the ‘Deploy’ tab in the RStudio console (bottom of IDE) will indicate sucessful deployment and the app should appear on shinyapps.io with a valid and public URL.
Debugging
Debugging can be tricky in Shiny apps due to the fact that the code is not run sequentially and you aren’t actually running the code yourself, it’s run in the background as the app is running. Especially once you have a lot of reactivity finding the exact source of the bug can be challenging.
You’ll find the ways that work best for you, but two common strategies I use are the following: - print variables to the screen within an observe function or within specific reactives - use the browser function
Debugging
Here is an app that contains a couple simple bugs and opportunities for optimization where these strategies can be helpful. This app exists as a file called ‘app_debug.R’ in the data folder of the course.
ui <- page_fluid(
fileInput("de_file", "Upload a DE file", accept = c(".csv", ".tsv", "xlsx", "xls")),
uiOutput("all_data_UI")
)
server <- function(input, output){
de_table_in <- reactive({
req(input$de_file)
rio::import(input$de_file) %>% dplyr::mutate(negLog10_pval = -log10(pvalue))})
output$all_data_UI <- renderUI({
req(de_table_in)
card(card_header("All genes"), dataTableOutput(outputId = "all_data"))
})
output$all_data = renderDataTable(datatable(de_table_in()))
}
shinyApp(ui = ui, server = server)Debugging
Two bugs are apparent:
Upload error - print to screen
Usually bugs with reasonable errors printed to the console are easier
to debug, so we will start there. Here is the error form the console:
If you think you know the variable causing the bug, then printing it
to the screen is one way to peek inside the app and look at it. We can
tell from the error in the console that it’s occurring in the
de_table_in reactive and some kind of issue with a ‘file’
argument. It must be a problem with the command
rio::import(input$de_file)
Upload error - print to screen
You can use the observe function to print a reactive variable, like input$de_file, and every time it changes, the observer will run and the variable will print to the screen.
# UI object not shown
server <- function(input, output){
observe({
print("file_path") #<<
print(input$de_file) #<<
})
de_table_in <- reactive({
req(input$de_file)
rio::import(input$de_file) %>% dplyr::mutate(negLog10_pval = -log10(pvalue))})
output$all_data_UI <- renderUI({
req(de_table_in)
card(card_header("All genes"), dataTableOutput(outputId = "all_data"))
})
output$all_data = renderDataTable(datatable(de_table_in()))
}
shinyApp(ui = ui, server = server)Upload error - print to screen
From the console, we can see that the variable input\(de_file* is a dataframe, not a file path like we need for this command. We then remember that we actually want *input\)de_file$datapath.
Upload error - print to screen
We could also put the print function within the specific reactive context where we think it breaks, in this case the de_table_in() reactive.
Sometimes you might want to know the value of input$de_file only within this reactive and this would be the best strategy, and sometimes you might want to track how its changing as the user goes through the app, and printing within an observe function might be more useful.
# UI object not shown
server <- function(input, output){
de_table_in <- reactive({
req(input$de_file)
print("file_path") #<<
print(input$de_file) #<<
rio::import(input$de_file) %>% dplyr::mutate(negLog10_pval = -log10(pvalue))})
output$all_data_UI <- renderUI({
req(de_table_in)
card(card_header("All genes"), dataTableOutput(outputId = "all_data"))
})
output$all_data = renderDataTable(datatable(de_table_in()))
}
shinyApp(ui = ui, server = server)Upload error - browser
The browser in R, called with the function browser() is a really useful way to debug functions and Shiny apps. In both cases, code is being run within a contained environment without you running it line by line.
By inserting browser() into an app or function, the code execution will pause, allowing you to manually run code within the environment of the app. So any temporary variables available to that function or Shiny app can be viewed.
Note: the browser won’t work properly if you are running this within an R markdown chunk, so open up the demo app (‘app_debug.R’) and run the app using the ‘runApp’ button at the top right.
Upload error - browser
We insert a browser function call within the de_table_in reactive, and when we run this (run in separate script), after a file is uploaded we can go into the console and run code from within the app environment.
# UI object not shown
server <- function(input, output){
de_table_in <- reactive({
req(input$de_file)
browser() #<<
rio::import(input$de_file) %>% dplyr::mutate(negLog10_pval = -log10(pvalue))})
output$all_data_UI <- renderUI({
req(de_table_in)
card(card_header("All genes"), dataTableOutput(outputId = "all_data"))
})
output$all_data = renderDataTable(datatable(de_table_in()))
}
shinyApp(ui = ui, server = server)Upload error - browser
By printing input\(de_file* within the browser, its clear that we need to change it to *input\)de_file$datapath
uiOutput error - browser
The second bug is trickier because there is no error. A version of this app with the upload bug from above fixed is in the ‘data’ folder of the course materials called ‘app_debug2.R’. Run this script with the ‘Run App’ button to see the bug.
uiOutput error - browser
We don’t have an error to pinpoint where in our app the bug exists, but since the card is appearing inappropriately, we can assume it’s an issue in the renderUI function that renders the card.
Similar to the last bug, we can insert a browser function call within renderUI and every time that funciton is run, we will know and can explore the issue.
# UI object not shown
server <- function(input, output){
de_table_in <- reactive({
req(input$de_file)
rio::import(input$de_file$datapath) %>% dplyr::mutate(negLog10_pval = -log10(pvalue))})
output$all_data_UI <- renderUI({
req(de_table_in)
browser() #<<
card(card_header("All genes"), dataTableOutput(outputId = "all_data"))
})
output$all_data = renderDataTable(datatable(de_table_in()))
}
shinyApp(ui = ui, server = server)uiOutput error - browser
In the browser we are able to look at the de_table_in variable we have used in req() a d notive that it’s a function, not a table as we intended.
It’s clear now that we forgot to add the parenthesis required to get the actual value from the reactive. We have returned the reactive function itself. Because this is not null, the req() function was satisfied and the card was rendered.
The old line of code req(de_table_in) should be changed
to req(de_table_in()). ]
]
Further Resources
Contact
Any suggestions, comments, edits or questions (about content or the slides themselves) please reach out to our GitHub and raise an issue.
Time for an exercise!
Exercises for Session 3 are here